In traditional classification schemes, the universe of knowledge is brokeii down into self- contained disciplines which are further analysed to the point at which a particular concept is located. This leads to problems of:

(a) currency: keeping the scheme in line with new discoveries.

(b) hospitality: allowing room for insertion of new subjects

(c) cross-classification: a concept may be considered in such a way that it fits as logically into one discipline as another.

Machine retrieval is also hampered by the fact that any individual concept is notated differently, depending on where in the scheme it appears.

The approach now considered is from an organized universe of concepts, every concept being set down only once in an appropriate vocabulary, where it acquires the notation which identifies it wherever it is used. It has been found that all the concepts present in any compound subject can be handled as though they belong to one of two basic concept types, being either Entities or Attributes. In classing, these concepts are identified, and notation is selected from appropriate schedules. Subjects are then built according to formal rules, the final class number incorporating operators which convey the fundamental relationships between concepts.

From this viewpoint, the Rules and Operators of the proposed system can be seen as the grammar of an IR language, and the schedules of Entities and Attributes as its vocabulary.